search

ohhhyeahhh / SiamGAT

Code for the paper "Graph Attention Tracking". (CVPR2021)
https://openaccess.thecvf.com/content/CVPR2021/papers/Guo_Graph_Attention_Tracking_CVPR_2021_paper.pdf
139 stars 19 forks source link

About the training datasets? #7

Open Tchuanm opened 3 years ago

Tchuanm commented 3 years ago

Thanks for your great work. I noticed that SiamGAT is trained in 5 datasets. While the baseline SiamCAR is trained in 4 datasets except for GOT10K. Have u compared the difference between these two training methods on performance? Looking forward to your reply. thanks advance.

twotwo2 commented 3 years ago

The diversity of datasets can really help the model to have better generalization ability. We also try to train SiamCAR in 5 datasets, but the improvment effect is small. I guess the four datasets already have enough diverse training samples.